- Email: [email protected]
Passenger evacuation simulation considering the heeling angle change during sinking
International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Contents lists available at ScienceDirect
International Journal of Naval Architecture and Ocean Engineering journal homepage: http://www.journals.elsevier.com/ international-journal-of-naval-architecture-and-ocean-engineering/
Passenger evacuation simulation considering the heeling angle change during sinking Hyuncheol Kim a, Myung-Il Roh b, Soonhung Han a, * a b
Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, South Korea Department of Naval Architecture & Ocean Engineering, Seoul National University, Seoul, South Korea
a r t i c l e i n f o
a b s t r a c t
Article history: Received 28 October 2017 Received in revised form 4 June 2018 Accepted 25 June 2018 Available online xxx
In order to simulate the evacuation simulation of a ship during a sinking, the slope angle change of the ship must be reflected during the simulation. In this study, the passenger evacuation simulation is implemented by continuously applying the heeling angle change during sinking. To reflect crowd behavior, the human density and the congestion algorithm were developed in this research and the walking speed experiment in the special situation occurring in the inclined ship was conducted. Evacuation simulation was carried out by applying the experimental results and the change of the walking speed according to the heeling angle of the ship. In order to verify the evacuation simulation, test items suggested by International Maritime Organization (IMO) and SAFEGUARD Validation Data Set conducted on a large Ro-PAX ferry (SGVDS 1) which performed real evacuation trial in full-scale ships were performed and the results of simulation were analyzed. Based on hypothetical scenario of when a normal evacuation command is delivered to the passengers of MV SEWOL in time, we predicted and analyzed the evacuation process and the number of casualties. © 2018 Production and hosting by Elsevier B.V. on behalf of Society of Naval Architects of Korea. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Keywords: Advanced evacuation analysis Heeling angle change Walking speed experiment MV SEWOL UNITY
1. Introduction Cruise passengers are expected to total 25.8 million in 2017 and have increased by 62% over the past decade (CLIA, 2016). As demand for cruise and passenger ships is expected to increase steadily, safe ship design and predictions of disaster situations are essential. In the case of these cruises or passenger ships operating at sea, there are usually many people on board and their evacuation route is narrow and complicated. In addition, when the sinking occurs, the slope is changed and the evacuation time is limited. Also, due to the difficulties in rescue, it can cause serious casualties in the event of a disaster. Therefore, accurate evacuation analysis is needed that reflects the special environment of marine environment that is stricter than the buildings onshore. IMO MSC.1/Circ. 1533 (IMO, 2016) proposes a method of using computer simulation that reflects the characteristics of individual passengers. Although it is not mandatory, the simulation result is recognized as a recommendation item and safety of passengers is secured by calculating the time required for the passenger to evacuate in an emergency situation and
* Corresponding author. E-mail address: [email protected] (S. Han). Peer review under responsibility of Society of Naval Architects of Korea.
identifying and eliminating the congestion and risk factors. In this research, to accurately reflect the sinking situation on simulation, the heeling angle change with time was applied continuously. Human density algorithm and congestion algorithm developed in this study were applied to reflect crowd behavior. Also, we applied the experimental results on the walking speed in a specific situation occurring in an inclined ship. For the verification of the simulation, the test items proposed by the IMO and the Safeguard validation data set-SGVDS1, which is an experiment on the actual ship (Galea et al., 2012), were performed and simulation results were analyzed whether the results met the criteria. By using the developed techniques in this research, the evacuation simulation for the scenario of three angles was performed and analyzed for MV SEWOL. In order to compare and analyze the evacuation status and time according to the heeling angle, the simulation was performed about the situation when the accident is not occurred at a heeling angle of 0 , the heeling angle of 30 about 8:50 AM and finally the heeling angle of 52.2 about 9:34 AM. Using the simulation results, we predicted and analyzed the evacuation process and the number of casualties based on the hypothetical scenarios when the normal escape command was delivered in time. Applying the processes and results of this research to other passenger ships, it is possible to design a ship with high reliability of safety by
https://doi.org/10.1016/j.ijnaoe.2018.06.007 2092-6782/© 2018 Production and hosting by Elsevier B.V. on behalf of Society of Naval Architects of Korea. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
2
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
confirming the evacuation possibility of passengers for various disaster situations and modifying and supplementing the congestion point. In addition, we expect to be able to minimize the casualties in case of an actual emergency situation by understanding the optimum measures for the accident situation in advance. 2. Related works The evacuation simulations for ships were mainly carried out since the early 2000s, following the study of buildings onshore. The most important factor in the evacuation simulation is to express the behavior of the passenger and the external environments that affect passengers. Walking speed is one of the most important factors in passenger behavior. Walking speed depends on passenger's age, gender, height, weight, size, disability, destination, environment, risk factors, awareness time, etc. The simulation results vary greatly depending on walking speed. Agent-based method is able to reflect individual characteristics. The agent-based method uses the coordinate system to move passengers and thus it is possible to indicate the exact position of each passenger. However, agent-based method requires immense computational capacity and time. In recent years, an activity-based method has been used in combination with other existing methods to express behaviors in specific situations (Santos and Aguirre, 2004). It is also important to reflect not only individual behavior, but also crowd behavior and emergency behavior in evacuation simulation that large numbers of people move. Cohesion behavior to move to a group location, Alignment behavior to move along the direction that the group moves and separation behavior to keep a certain distance between passengers in a group are considered as representative group behavior in evacuation simulation (Reynolds, 1987). Table 1 shows a comparison of representative methods and features of ship evacuation simulation researches that have been performed so far. Evi is a representative simulation software of passenger ship evacuation analysis that expresses cohesion and alignment behaviors by moving at the same walking speed between groups. The separation behavior is expressed by a method of calculating the number of passengers in a certain space and reducing the walking speed when the density is high (Vassalos et al., 2002). FDS þ EVAC is specialized for smoke and fire propagation and is mainly used for evacuation analysis in case of fire in a closed space such as a building or a turnnel (Korhonen and Hostikka, 2009). maritime EXODUS is a passenger evacuation analysis program using velocity based fine network model. The passenger uses each cell that stored the distance from the starting point to the arrival point to move in the direction of decreasing the distance value (Gwynne et al., 2003). Pathfinder has a SFPE mode that reflects the criteria presented by the U.S Fire Technology Society. However, it does not reflect the situation in which slope of ship is continuously changed as other softwares.
ODIGO is a program for marine specific, using artificial intelligence to express the behavior of passengers (Pradillon, 2004). The simulation of this study was implemented using the UNITY engine. UNITY engine is a game development platform that adds Graphical User Interface (GUI) to Application Programming Interface (API). It has high visualization quality and freedom of development, and can describe various physical characteristics and environment (UNITY, 2008). The previous researches that applied equally reduced walking speed during simulation by applying the fixed slope at a certain time point despite the importance in the ship evacuation simulation. In this research, we apply the change of slope to the passenger continuously by recalculating the walking reduction factor according to the angle during the simulation. We also applied the human density and congestion algorithm developed in this research to simulation to reflect human characteristics more realistically. One of the most difficult parts of the evacuation simulation is verification and validation of the results. Not only is there not enough data for verification, but it is not possible to experiment with the actual scale of an emergency situation that has to escape, so in many cases it depends on assumptions and conjectures to analyze the simulation results. Using the MIXAMO Engine (MIXAMO, 2017) that provides skeletons and animations to 3D models, realistic motion is visualized for passengers. It will be helpful to understand and evaluate the evacuation situation in the result analysis. 3. Modeling method and algorithm For high accuracy evacuation simulations, it is important to use algorithms that reflect actual human characteristics and describe detailed disaster environment. This chapter introduces the techniques applied in this paper, describes the implementation method and the applied algorithm. 3.1. Walking speed reduction factor due to slope of a ship Walking speed is one of the most important factors in evacuation simulation. Passengers are greatly influenced by individual characteristics, crowd density and walking environment. In the case of a ship, passengers may have to evacuate from an inclined ship as well as the normal situation. This situation make the movement more difficult than general situation and causes a walking speed reduction and greatly lengthens the total evacuation time of passengers. Therefore, it is necessary to consider the walking speed change according to the slope when performing the evacuation simulation about the ship. Typically, it is assumed that a person cannot walk at the slope greater than 35 due to the loss of friction. Therefore, most of the studies have been performed only at the slope below 35 (Cho, 2011). Lee et al. (2003) compared and analyzed the walking speed experiments when the trim and
Table 1 Technical comparison of previous works with our work. Gwynne et al. (2003) Ginnis et al. (2010)
Azzi et al. (2011) Park and Lee (2016) Ha et al. (2013) Zhang et al. (2017) This work
VELOS
Evi
Pathfinder
In-House
FDS þ EVAC
UNITY
Agent model Agent rotation Account for slope
maritime EXODUS Velocity X Constant
Velocity B Constant
Velocity X Constant
Velocity B Constant
Acceleration B Constant
Acceleration B None
Method considering slope
Reduction factor
Projection of gravity force Reduction factor Reduction factor
External force X △
Reduction factor X △
Velocity B Continuous/ Prediction equations Reduction factor B B
Simulation platform
Account for walking direction X Visualization △
X B
X △
X B
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
3
heeling angle were present. Cho (2011) applied a regression analysis to the walking experiment results to find the walking speed reduction factors according to slope and the reduction factors are obtained about three situations where the flat ground is inclined and five situations where the stair is inclined. In this research, the simulation was implemented using the results of regression analysis on the walking test results of Cho (2011). Fig. 1 shows each situation. Fig. 2 is the plot of each estimation equation about reduction factor. At 0 without slope, the reduction factor is 1 and there is no decrease in walking speed due to the slope. However, the reduction factor reaches 0 because it is assumed that walking is impossible at the slope greater than 35 . Also, reduction factors for all situations is equal to 0 at 1 and 35 at 0, but the reduction factors according to each situation are different even though the slope is same at the angle between 0 and 35 . In this research, evacuation simulation was performed using the reduction factor to apply the walking speed change according to the slope of the ship. Fig. 2. Plots of walking speed reduction factor.
3.2. Walking speed experiment in corridor of inclined ship In a situation where there is a slope like a ship, it is possible to increase the accuracy of the simulation by applying the walking speed in special situation as well as the walking speed change due to the slope. When there is no slope, the corridors and walls turn into walls and corridors when the slope is present, respectively. A passenger may have to walk using a wall that has changed from wall to corridor due to the slope. In order to apply the walking speed for this special situation to the simulation, a simple experiment was conducted on two 24 years old women using a truck capable of creating an inclined environment (Oh, 2017). Fig. 3 represents the experimental scene. Experiments were conducted on four situations, such as walking with one leg on the floor and another on the wall (walking with wall and floors), rolling sideways on a floor or wall (rolling
sideways), walking using the corner between the floor and the wall (corner walking), and moving while putting two feet on the floor and hands on the wall (walking sideways with hands and feet) that can occur in a situation when the heeling angle is present. Table 2 shows the experimental results for four situations. In IMO MSC.1/Circ. 1533, the minimum walking speed of females younger than 30 years in flat terrain is 0.93 m/s and in stairs up and stairs down are 0.47 m/s and 0.56 m/s, respectively. The minimum speed while climbing stairs is similar to the speed while walking with one leg on the floor and another on the wall (walking on wall and floor). The consideration here is that forces received by a human body due to posture while walking on wall and floor, is similar to the one while climbing stairs. In case of rolling sideways, walking sideways and walking through corners, low movement speed was
Fig. 1. Walking cases in inclined ship.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
4
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 3. Experiment on special situations.
Table 2 Experiment results on special situations.
Moving speed (m/s)
Walking with wall and floors
Roll sideways
Corner walking
Side walking using hands and feet
0.469
0.347
0.303
0.322
measured, since such movements are difficult comparing to regular ones. This experiment is intended to provide an approximate estimation of the walking speed in special situations and has limitations with too few participants. In order to achieve more accurate results, we will conduct experiments later with varying gender and age group.
3.3. Heeling angle change during sinking The ship that is sinking changes its slope according to time. In order to simulate the accurate and realistic evacuation of the passenger ship, the evacuation simulation should be performed by applying the slope varying with time to the walking speed of passenger instead of applying fixed walking speed according to the slope at a specific time. Passenger ships are more complex than buildings, so passengers will go through all route states of the transverse ground and the upward and downward ground during the evacuation. Even though the slope is the same, the walking speed reduction factor varies depending on the route state. Therefore, it is necessary to know the current route status of the passenger in order to apply the correct reduction factor to the evacuation simulation. Fig. 4 shows the walking situation that can be occurred in a ship having a heeling angle. Since the simulation proceeds according to the discrete time interval, it is possible to obtain dyp by comparing yp n that position of current time step and yp n1 that the position of previous time step in the y-axis coordinate. By using dyp , the current path condition can be recognized. Eq. (1) represents the discriminant used in this study to determine the path condition.
yp n yp n1 ¼ dyp
8 < ¼ 0 : Transversal > 0 : Upward : < 0 : Downward
(1)
If dyp that is difference between yp n and yp n1 is 0, it means that the transversal ground is used because y-axis position does not change. If dyp is larger than 0 or smaller than 0, it means that the upward or downward ground is used, respectively. The evacuation simulation performed in this research is agent-based method and agent-based method represents the position of passengers in a coordinate system. Using the algorithm proposed in this research, it is possible to easily recognize the current path condition and apply the walking speed reduction factor according to the slope and path conditions of the ship.
3.4. Human density algorithm The walking speed of passengers depends on the number of nearby passengers. The walking speed in the case in which the passenger density is small and the passengers do not influence each other is referred to as a free walking speed, and in case of a crowd, the walking speed is slower than the free walking speed due to the influence of other passengers. As the number of passengers nearby increases, i.e., as the passenger density increases, the walking speed decreases. This phenomenon is due to the separation behavior to maintain a certain interval between passengers during the group movement. In the evacuation simulation, it is practical to carry out the simulation by taking into consideration both the walking speed when the group moves instead of free walking speed, and the
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
5
Fig. 4. Route state in a ship with heeling angle.
walking speed when encountering other passengers coming from the opposite direction. At this time, the factor that affects the walking speed is the passenger density, which refers to the number of passengers per area. Lee et al. (2003) compared and analyzed the results of walking speed experiment according to population density. Nelson and Mowrer (2002) reported that a free walking speed is possible until 0.54 person=m2 but walking speed at 3.8 person=m2 or more is significantly slower. In the research of Togawa (1955), the walking speed was reduced to about 0.1 m=s by 12 times when the group moves horizontally at 6.0 person=m2 or more. In IMO MSC/Circ.909 (IMO, 1999), the walking speed according to passenger density is divided into low, optimum, moderate and crush. EVACNET4 (Kisko et al., 1998), an evacuation simulation software, divides the passenger density into six stages according to unit area per person (A: More than 1.2, B: 0.93e1.2, C: 0.65e0.93 D: 0.25e0.65, E: 0.19e0.28, F: Lower than 0.19) and reduces the speed according to stages. However, when calculating the density of human to calculate walking speed, it is not realistic to calculate the occupancy of passengers in a certain space. Instead, when calculating the passenger's walking speed, it is more reasonable to consider only the other passengers who enters the field of view and directly affected the movement of the passenger in which the calculation takes place and ignores the passengers behind the moving direction. In this study, Ray-casting (Roth, 1982) method was used to consider only the passengers directly affecting the movement in density calculation. Ray-casting is a method of projecting rays to cross-check with a specific object. Table 3 shows the comparison of the passenger density method between area occupancy method and this research method. The results of Togawa (1955) and a quadratic function were used to estimate the walking speed reduction factor. Fig. 5 is a plot of the estimation equation used in this research. The minimum walking speed reduction factor is given as 0.08 that is 12 times lower than the speed reduction factor when free walking. During the simulation, the presence of nearby passengers affecting the walking speed is checked at every discretized time step and if there are passengers, the number of passengers is counted. The walking speed reduction factor is calculated from the number of passengers and applied to the walking speed individually set at the start of the simulation to move the passenger at a reduced walking speed according to the number of nearby passengers. In order to examine the feasibility of the proposed human density algorithm in this research, the method without considering human density, the area occupancy method, and a human density algorithm that proposed this research for 100 passengers were compared. Fig. 6 shows the results of these simulations according to time. The maximum walking speed of all passengers was determined by arbitrarily distributing a value of males 30e50 years old
(0.97e1.62 m/s), suggested by IMO MSC.1/Circ. 1533. The result when the passenger density was not taken into consideration showed an unrealistic phenomenon that cannot occur when moving the group. This is only possible when passengers have been trained in advance. Furthermore, walking speed without considering human density did not take into account the influence of other people, which did not reflect the separation phenomenon to maintain a constant spacing among the crowd behavior. In the case of the area occupancy method, the influence of passengers located behind and beside the calculating passenger is considered. Therefore, an excessive walking speed reduction occurs more often than necessary. As time passes, the gap between the passengers who are located in front of the moving direction, who are less affected by other passengers, and the passengers located at the back becomes wider. An unrealistic phenomenon has occurred in which the passengers who are located at the rear are moved together. In addition, the passengers who are performing the calculation show unreasonable results because they consider the passengers behind the moving direction, and move at a reduced speed. The walking speed of this research differs greatly between the front and the rear passengers in the same way as the area occupancy method at the beginning of the simulation. However, as the simulation progresses, the gap is gradually reduced and it is confirmed that the distance between passengers is maintained without moving together. This study confirms that passengers are affected by other passengers within the cognitive area and the crowd behavior varies according to the range and characteristics of the cognitive area. It is more realistic to consider only a certain range of passengers entering the field of vision rather than considering all passengers in a particular area including the rear of the moving direction. 3.5. Path finding & congestion algorithm In this evacuation simulation, the calculation of path finding was performed using the A* algorithm (Hart et al., 1968) and the collision avoidance between passengers was performed using the Reciprocal Velocity Obstacles (RVO) (Snape et al., 2012) model. The A* algorithm is a graph/tree search algorithm that finds the shortest path from a given point to final destination. The cost is estimated by evaluating the heuristic estimate. Eq. (2) shows the cost estimation method of the A* algorithm.
f ðnÞ ¼ gðnÞ þ hðnÞ
(2)
f is the total cost, expressed as the sum of g and h, and n is the point of departure. g represents the moving cost from the starting point to the moving point, and the value of g increases as the calculation proceeds. h is a Heuristic function which means the moving cost from the moving point to the final destination. The Manhattan
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
6
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 3 Comparison between area occupancy and this work method.
Fig. 5. Walking speed reduction factor for varying calculated no. of neighbor agent.
method is used in which the obstacle and the diagonal direction are ignored and only the lateral and transverse direction is considered. Fig. 7 shows the process of reaching and calculating the arrival cost from the starting point to the destination using the A * algorithm. The RVO model was developed to overcome the vibration phenomenon of the Agent, which is a disadvantage of the Velocity Obstacle (VO) model, and to give the agent more realistic behavior. In the VO model, agents that are computed assuming that others
are obstacles are responsible for all collision avoidance. In the RVO, unlike the VO model, it is assumed that the other person is not the obstacle and all nearby agents are responsible for collision avoidance. Also, RVO can represent smoother motion than VO model. In addition, in this research, the characteristics of human was reflected at congestion points to enhance reality. To reflect human characteristics as realistically as possible, we analyzed and defined human thought patterns of congestion points. Preferentially, a step to recognize the congestion is needed. It is assumed that even though the same situation is given according to the personality and tendency of an individual, it is different for each person to judge and perceive as a congestion. Once the congestion point is identified, it is assumed that passengers will think or confirm about other available routes and rethink about whether to actually change the route or keep the current route even if other available routes are checked. In order to apply human characteristics to evacuation simulation algorithms, we define four variables (CA , CT , CD and CW ) representing human tendency were defined. Fig. 8 shows the algorithm at the congestion point applied to the evacuation simulation of this research. CA is the congestion agent number that represents the number of passengers existing within a specific forward range of passengers, for whom the calculation is being performed. CT is the congestion thinking number. Each passenger is assigned a different CT value, which represents the tendency of a person. In general,
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
7
Fig. 6. Comparison of simulation results according to density algorithm.
Fig. 7. Process of A star algorithm for path finding.
Fig. 8. Logic of congestion algorithm in simulation.
people show different tendencies, even though they are in the same condition. Only when C_A > C_T, a passenger recognizes that there is a congestion. The congestion decision number, C_D, and congestion waiting number, C_W, are related to the changing of route at the last stage. Each passenger is also assigned a different C_D and C_W value, as they are related to the personality of a passenger. We assume that, even if the passengers are aware that the current path is congested, they will also be reluctant to change the path. Only when C_D > C_W, the passengers will actually change the route and proceed to evacuate. In order to change the route during the simulation, first of all, it is necessary to exclude route recognized to be congestion and re-search the route. This process is implemented by using bit mask. The bit mask is to assign a mask such as 0 or 1 to a bit. Each corridor is assigned an eigenvalue increasing by 2n and the decimal value obtained by using route discriminant shown in Eq. (3) is converted into binary numbers. Each passenger can recognize the congestion
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
8
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 4 Eigenvalue according to corridor number. Bit
∙∙∙
LSB-5
LSB-4
LSB-3
LSB-2
LSB-1
LSB
Corridor number
∙∙∙
5
4
3
2
1
Eigen value
2n
32
16
8
4
2
Route State 0 or 1
phenomenon of the route through the bit mask. Table 4 shows the eigen value and bit according to the route. The Least-Significant Bit (LSB) indicates the least-significant bit that located at the end of the bit configuration. An example of some cases is shown in Table 5. Route discriminant ¼ Path state þ Eigen value
(3)
If Routes 1, 3, and 5 are detected as congestion corridors, this will result in a decimal value of 43 by the route discriminant equation, and 101011 when converted to binary numbers. Conversely, the first bit indicates the existence of the route itself, and the second bit indicates the corridor number in order. A value of 1 in the route indicates that the route is a congestion and should be changed and a value of 0 indicates a route that is not a congestion or an unidentified route. When each occupant judges that the current route is unavailable due to the congestion, the route re-search must be performed again, and the evacuation should proceed using the other route. It is possible to search for the route again, except for the unavailable routes that are recognized by using the bit mask. Fig. 9 shows the simulation results for the behavior of the passengers at the congestion point. 4. Validation/verification of evacuation simulation IMO established MSC.1/Circ. 1533 to acknowledge the evacuation time derived from computer simulations. In addition, guidance on validation/verification of evacuation simulation tools was presented in ANNEX 3, and the tests for verifying whether a reasonable evacuation simulation can be performed is presented. Tests 1e7 are component testing that involves checking that the various components of the software perform as intended, and tests 8e12 are qualitative verification that concerns the nature of predicted human behavior with informed expectations. The evacuation simulation implemented in this research performed the tests presented in IMO MSC.1/Circ. 1533 for validation and verification. It was confirmed that all 12 tests requirements were satisfied. In this chapter, to demonstrate that the test requirements are satisfied, the results of tests 8, 10, and 12 that are more complex qualitative verification tests rather than the relatively simple component tests, are presented. In addition to the IMO test, the SAFEGUARD validation data set 1 was used for further verification and the results were analyzed. 4.1. IMO test 4.1.1. Test 8: counterflow Test 8 takes place in a space where two rooms, 10 m wide and long, are connected via a corridor 2 m wide and 10 m long. In the first room, 100 passengers move through the corridor to the second Table 5 Example of route status. Unavailable route
Route discriminant
1 5 1,3,5 2,4,6
1 1 1 1
þ þ þ þ
2¼3 32 ¼ 33 2þ8 þ 32 ¼ 43 4þ16 þ 64 ¼ 85
Binary number 11 100001 101011 1010101
Fig. 9. Simulation of congestion (MV SEWOL A deck).
room. When the passengers in the first room move to the second room, the time when there is no passenger coming from the opposite side and when 10, 50, and 100 passengers are coming from the opposite side are measured. The expected result is that as the number of passengers coming from the opposite side increases, the time required for the passengers in the first room to reach the second room increase. Fig. 10 shows the simulation results for the case where the counterflow is absent, and the case where the counterflow increases at 10, 50, and 100 passengers, respectively. It is confirmed that as the amount of counterflow increases, the time required to move from the first room to the second room increases. Table 6 shows the results of comparison with other research results for additional verification. 4.1.2. Test 10: Exit route allocation Test 10 takes place in a space where a total of 23 passengers are distributed among 12 rooms. The expected result is that the allocated passengers move to the appropriate destinations. Fig. 11 shows the result of test 10. It was confirmed that all the passengers move to the appropriate destinations along the corridor without penetrating the wall. 4.1.3. Test 12: flow density relation Test 12 is for demonstrating that the flow of persons in the corridor is generally smaller at very high population densities
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
9
Fig. 10. Simulation result of test 8.
Table 6 Comparison of simulation results. No. of counterflow
Cho (2011)
Evi
This research
88.9 () 125.6 (41.3%) 229.1 (157.7%) 327.9 (268.8%)
82.3 () 108.1 (31.3%) 224.3 (172.5%) 277.7 (237.4%)
(sec) 0 10 50 100
84.6 () 93.2 (10.2%) 137.1 (62.1%) 216.1 (155.4%)
compared with that at moderate densities. From the same starting point, 20, 50, and 70 passengers were deployed to conduct the test, and the simulation results are shown in Fig. 12. The flow of persons was calculated by measuring the time taken for the last passenger to pass through the 30-m corridor. Table 7 shows the results according to each number of passengers. It was confirmed that as the number of passengers increased, the flow of persons decreased. This phenomenon can be regarded as a result of the separation behavior among the crowd behaviors to maintain the minimum distance between passengers.
4.2. SAFEGUARD validation data set 1 SAFEGUARD Validation Data Set 1 (SGVDS 1) was conducted on an actual Large RO-PAX ferry, with a total of 1349 passengers
boarded. There are four metrics to validate the simulation. The first is Euclidean Relative Difference (ERD), as shown in Eq. (4). This is used to assess the distance between the experimental data ðEi Þ and the model data ðmi Þ. If ERD is 0, this means that the two data are identical.
Em ¼ jEj
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1 ðEi mi Þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1 Ei
(4)
The second is Euclidean projection coefficient (EPC), as shown in Eq. (5). The EPC calculates a factor which when multiplied by each model data point ðmi Þ reduces the distance between the model ðmÞ and experimental ðEÞ vectors to its minimum. If EPC is 1.0, it implies that the difference between the model ðmÞand experiment ðEÞ data is as small as is possible.
Pn E; m Ei mi ¼ Pi¼1 n 2 2 m i¼1 mi
(5)
The third is the Secant cosine, as shown in Eq. (6). It provides a measure of how well the shape of the model data curve matches that of the experimental data curve. If SC is 1.0, it means that the shape of the model ðmÞ curve is identical to that of the experimental ðEÞ curve.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
10
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 11. Simulation result of test 10.
Pn
ðEi Eis Þðmi mis Þ i¼sþ1 s2 ðti ti1 Þ
E; m ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Em Pn ðEi Eis Þ2 Pn ðmi mis Þ2 i¼sþ1 s2 ðti ti1 Þ
(6)
i¼sþ1 s2 ðti ti1 Þ
The last parameter is the percentage difference between the predicted Total assembly time (TAT) and the measured TAT. Eq. (7) indicates TAT.
%TAT ¼ ðMeasured TAT Predicted TATÞ*100 = Measured TAT (7) The ship used for validation consists of three decks and Fig. 13 shows modeling results. According to SGVDS 1 guidelines, a total of 50 simulations were carried out while changing the distribution of the population in every fifth simulation as stated in IMO MSC/Circ. 1533. Table 8 shows Total Assembly Time (TAT) and ERD. The simulation of the smallest ERD should be selected for the basis of the validation
comparison. In this case, the values of case no. 11 are selected for analysis about two phase assessment process. The value calculated for case no. 11 are listed in Table 9. The suggested two phase assessment process in the SGVDS 1 guideline are as follows: - Phase 1: For the predicted total assembly curve, determine ERD, EPC, SC and % TAT. 1. ERD 0.45. 2. 0.6 EPC 1.4. 3. SC 0.6 with s=n ¼ 0.05. 4. Predicted TAT for the overall assembly to be within 45% of the measured value. - Phase 2: for the predicted assembly curve for each of the four assembly stations, determine ERD, EPC and SC. At least 9 out of 12 criteria must be met SGVDS 1 to satisfy the criteria and it is not acceptable to have two or more failed criteria in any one assembly station. The simulation results for SGVDS 1 in this study confirm that both Phase 1 and Phase 2 presented in the guideline are satisfied. 5. MV SEWOL evacuation simulation 5.1. System architecture The simulation in this research is divided into the PreTable 7 Comparison of flow of persons.
Fig. 12. Simulation result of test 12.
No. of agent
Density
Time ðsecÞ
Flow of persons ðm=sÞ
%
ðp=m2 Þ
20 50 70
0.95 2.38 3.33
57.78 94.48 112.28
0.63 0.38 0.33
e 39.68 47.62
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
11
Fig. 13. Geometry model of SGVDS 1 and distribution of passengers.
Table 8 TAT and ERD for SGVDS1 simulation. No.
Time
ERD
No.
Time
ERD
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
440.8 424.1 432.5 403.6 428.7 416.2 440.4 430.6 435.5 438.7 422.3 430.5 411.6 375.2 393.4 440.0 390.3 440.0 390.3 415.2 431.8 437.4 386.3 411.9 393.9
0.392 0.414 0.402 0.447 0.395 0.421 0.403 0.415 0.408 0.412 0.371 0.391 0.399 0.427 0.412 0.389 0.426 0.408 0.418 0.397 0.406 0.399 0.435 0.415 0.402
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
428.9 411.4 442.3 384.9 413.6 449.3 421.7 435.5 399.3 411.1 405.2 443.4 576.4 509.9 386.6 418.7 402.6 429.2 409.4 419.3 402.6 420.9 411.7 416.3 441.8
0.405 0.403 0.397 0.432 0.414 0.422 0.432 0.427 0.413 0.410 0.411 0.433 0.442 0.436 0.392 0.437 0.414 0.430 0.400 0.414 0.391 0.385 0.388 0.406 0.423
Table 9 Validation metric values of smallest ERD case for SGVDS 1.
Overall AS A AS B AS C AS D
ERD
EPC
SC (s/n ¼ 0.05)
%diff. TAT
0.371 0.335 0.381 0.256 0.668
0.897 1.190 0.909 0.977 0.630
0.877 0.698 0.835 0.820 0.706
27.81 43.28 28.37 22.93 30.10
Processing stage and the Run Evacuation Simulation stage. Fig. 14 shows the whole simulation process. In the Pre-Processing stage, analysis of the sinking of the ship for application to the simulation environment is performed and the initial information necessary for running the simulation is defined. The analysis of the sinking of the ship is transferred to Motion of Ship in the Run Evacuation
Simulation stage. The Definition of Evacuation Simulation is a step that takes all the overall matters and gives the passengers detailed characteristics (gender, age, body shape, athletic ability, personality, etc.) and assigns the initial position and the destination point. It also creates a navigation field for defining the movable space and applying the pathfinding algorithm. Each corridor is assigned an eigenvalue according to the bit in order to allow the passengers to cope with the congestion. When the Pre-Processing stage is completed, the Run Evacuation Simulation stage is executed. In the Global Navigation stage, the route is calculated from the initial position to the destination using the A* algorithm based on the information in the Pre-Processing stage. When the route calculation of all the passengers is completed, the Local Navigation stage is performed. In the Local Navigation stage, collision avoidance is performed between the passengers moving to the destination, and ray-casting is used at every time step to calculate the number of nearby passengers affecting each passenger's walking speed. The congestion algorithm is executed based on the calculated number of passengers. If the congestion condition is satisfied, the route is recalculated and the evacuation proceeds. In Motion of Ship stage, information on the slope from Ship Sinking Analysis stage and the current route condition of each passenger is obtained. In the Walking Speed Reduction Factor stage, the reduction factor is calculated based on the slope and route condition calculated in the previous stage and the number of nearby passengers that affect walking speed. The evacuation simulation is performed by repeating the Local Navigation stage by adding the discretized time until the passenger either is unable to move due to the slope of the ship or has reached the destination point. 5.2. Simulation scenario The simulation of evacuation simulation in MV SEWOL (Length: 145.6 m, Breadth: 22.0 m, Draft: 6.25 m, and Gross Registered Tonnage: 6825) was based on the report of the research institute and academic journal (KMST, 2014; Kim et al., 2016). Fig. 15 indicates the position and number of crews and passengers of MV SEWOL's Nav Deck (5th), A Deck (4th) and B Deck (3rd) according to the reservation record of the day of the accident. A total of 476 people boarded on the day of the accident, 17 on Nav Deck, 336 on A Deck and 123 on B Deck.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
12
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 14. Process of evacuation simulation.
Each simulation case was performed with awareness time of 5 min randomly assigned to the passengers since the time of the sinking accident occurred during the daytime and passengers start moving immediately without awareness time. The walking speed
of passengers was determined based on table 3.4 in IMO MSC.1/ Circ. 1533. Considering that most of the passengers were high school students, walking speed of man was applied by males younger than 30 years old that maximum walking speed is 1.85 m/s
Fig. 15. Deck layout of MV SEWOL.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
and minimum walking speed is 1.11 m/s. The walking speed of woman was applied by females younger than 30 years old that maximum walking speed is 1.55 m/s and minimum walking speed is 0.93 m/s. All crew members were assumed to be male and maximum walking speed and minimum walking speed are 1.85 m/ s and 0.93 m/s, respectively. The walking speed between the highest walking speed and the lowest walking speed of the corresponding group was randomly assigned to passengers to finally determine the passenger's walking speed in simulation. The recognition of the evacuation location and route was performed on the assumption that it was done in advance. The earliest possible time to make evacuation command is 8:50 AM at a heeling angle of 30 immediately after the accident. Based on an investigation report of the capsizing of the MV SEWOL that the heeling angle was 30 at 8:50 AM and 74 at 10:09 AM, the simulation was performed to reflect the heeling angle change under the assumption that MV SEWOL was linearly inclined as a 0.56 deg=min of heeling angular velocity. Because the stairs are difficult to use in a situation where the heeling angle of the ship is 30 , the tidal current at the accident spot is very fast and the water temperature is low, it was assumed that if there was an evacuation command, it would be more likely to induce the initial evacuation to the starboard deck that is the opposite direction to the sinking rather than to jump directly into the sea. The next possible time to make evacuation command is 9:34 AM at a heeling angle of 52.2 when the coast guard arrived and the crews were aware of the arrival of them. Normally, a person cannot walk at the slope greater than 35 due to the loss of friction. However, in the case of MV SEWOL, there were passengers who succeeded in evacuation at a slope of 52.2 or more. Therefore, evacuation simulations were performed under the assumption that passengers can grab structures and slip down. Since there is no experimental and simulation results for the slope of more than 35 , the walking speed was simulated by applying the mobility impaired (2), which is the slowest speed suggested by IMO. In the case of severe slope of the ship, the corridor becomes the wall and the wall becomes the corridor. We carried out the simulation using the experimental result of walking with wall and floors mentioned in section 3. In addition, in the situation where the heeling angle of ship is 52.2 , we assumed that it was possible that passengers can grab structures and slip down but it is impossible to climb up. Based on this assumption, simulation was performed by setting the port side that is the inclined direction as the evacuation point. To compare the simulation results of evacuation time, simulation that heeling angle of ship is 0 were also conducted to evacuate to the port and starboard side of each deck.
5.3. Evacuation simulation Table 10 shows the scenario and results of the evacuation simulation performed in this research and in Fig. 16, overall simulations are shown. The simulation video in this research can be found in Appendix A. Supplementary Video. Since the Total Evacuation Time (TET) is the most important result in this research, the results of TET are compared and TET are selected based on the convergence measure of total evacuation time (Ronchi et al., 2014). Eq. (8) represents the average value of TET when the simulation is performed n times and i is the i-th simulation run.
TETave ¼
n 1X TETi n
(8)
i¼1
A measure of the convergence of two consecutive mean total evacuation times is obtained calculating convergence of total evacuation time (TETconv ) and Eq. (9) represents TETconv .
TETave TETavn1 TETconv ¼ TET
13
(9)
ave
In this research, we use 1% of TETconv as our criterion. All simulation cases were repeated 50 times and TETconv of all results was less than 1% and confirmed that it met the criterion. 5.3.1. Case 1 (Heeling angle: 0 deg) Case 1 is simulation for evacuation time comparison. It was performed with a heeling angle of 0 , and the accident did not occur. Because there is no slope, the slope-dependent walking speed reduction factor is fixed to 1 during the simulation, and the simulation was performed only considering the walking speed reduction factor according to human density. When passengers evacuated to the port or starboard side, significant differences were not seen in the results. In the case of B deck, where the number of passengers is less than half that of A deck, the time required for evacuation is relatively long. The reason for this can be predicted as a result of the fact that B deck had many women, who were walking slower than men in the forward rooms that were farthest from the evacuation point. 5.3.2. Case 2 (Heeling angle: 30 deg) Case 2 is a situation that the heeling angle of ship is 30 at 8:50 AM, immediately after the accident based on hypothetical scenario of when evacuation command is delivered. The heeling angle of MV SEWOL increased at an angular velocity of 0.56 /min. It took only 535 s to reach 35 when the walking speed reduction factor becomes 0 due to the slope. Therefore, it is safe to successfully evacuate to the evacuation area in 535 s of ASET. Considering the awareness time, A deck took 439 s to evacuate the whole passengers. B deck was failed to evacuate the whole passengers and 97.5% of B deck passengers took 438 s to evacuate. Considering that after 535 s, movement of passengers was impossible, it was not a situation where passengers can evacuate easily. In the research of Park and Lee (2016), the time required for all passengers to evacuate to the port side of the B deck is 305 s at the heeling angle of 30 . Park and Lee (2016) performed a simulation using a walking speed reduction factor with a fixed heeling angle of 30 rather than reflecting the actual angular velocity during the sinking. For this reason, in comparison with the evacuation simulation of this research, it seems that even though it is more complicated and difficult situation such as more passengers and stairs are used, it took less time to evacuate. 5.3.3. Case 3 (Heeling angle: 52.2 deg) Case 3 is around 9:34 AM, just after the arrival of coast guard. At this time, the MV SEWOL was inclined at 52.2 . At a slope of 35 or more, it is common to assume that a person cannot walk because of the loss of friction. However, in the case of the MV SEWOL, some passengers succeeded in evacuating when the heeling angle was more than 52.2 . Therefore, evacuation simulations were performed under the assumption that passengers can grab structures and slip down. It is assumed that the walking speed is mobility impaired(2) when slipping down, and using the experimental result from walking with the floor and wall, described in Section 3, it is assumed that the walking speed is 0.469 m/s in a narrow corridor. In the case of B and A decks, it is impossible to evacuate through the port side due to the submergence according to heeling angle 12 and 15 min after 9:34 AM, respectively. Therefore, the simulations were performed assuming that the ASET of A and B decks are 15 min and 12 min, respectively. The simulation results of A deck show that 582 s with awareness time and 556 without awareness time. In case of B deck, the simulation results show that 429 s with awareness time and 227 s without awareness time.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
14
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 10 MV SEWOL evacuation simulation scenarios and results. Case
Time (AM)
Initial heeling angle (deg)
Deck
Evacuation point
Heeling angular velocity (deg/min)
Awareness time (sec)
ASET (sec)
RSET (sec)
% of evacuee
1e1 1e2 1e3 1e4 1e5 1e6 1e7 1e8 1e9 1e10 1e11 1e12 2e1 2e2 2e3 2e4 2e5 2e6 3e1 3e2 3e3 3e4 3e5 3e6
e
0
NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd)
Starboard side of each deck
0
300
No limitation
301.7 317.6 319.7 36.4 91.8 87.3 298.5 322.1 331.9 29.4 93.1 96.3 354.9 421.9 421.8 64.7 197.3 175.6 325.9 387.8 355.6 67.8 286.9 259.1
100
08:50
09:34
30
52.2
0
Port side of each deck
300
0
Starboard side of each deck
0.56
300
535
0
Port side of each deck
0
300
0
1200 900 720 1200 900 720
Fig. 16. Simulation examples: (a) Overall view of MV SEWOL (b) Passenger classification (c) forward of A deck (d) Aft of A deck (e) Overall view of A deck (f) Forward of B deck (g) Middle of A deck (h) Slipping motion (i) Middle of B deck (j) Walking corridor with wall and floors in inclined ship (k) Slipping motion in A deck.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
6. Conclusion In this research, for realistic evacuation simulation of passenger ships, the actual heeling angle change of MV SEWOL during sinking according to time was continuously reflected in the passenger's walking speed, and experimental results for special situations were used in simulation. Human density and congestion algorithms were developed and applied to reflect human characteristics. The simulation was carried out at the time when the evacuation command could be taken based on investigation report of the capsizing of the MV SEWOL. Simulation results show that if there was an immediate evacuation command at the time of the accident, all passengers were able to survive. However, at the time of the arrival of the coast guard, there were deaths, even if the evacuation command was delivered. Through this study, it can be seen that the responsibility and prompt response of the captain and the crew are very important factors which are directly connected with the casualties in case of ship accident. Normally, in the case of a slope of 35 or more, it is impossible for a human to walk. However, there may be situations where a human can walk using other structures or slipping down. It is expected that a more realistic simulation will be possible if data on various situations are acquired through experiments. By applying the implementation process and results of this research to other ships, it is possible to confirm evacuation performance and congestion points in advance and to make more realistic evacuation manual. Through such predictions about disaster situation, we expect to minimize the casualties in the event of an actual emergency. In the future research, we plan to study the evacuation simulation by adding the situation that affects the movement of passengers by moving the furniture or equipment due to the slope of the ship and selecting the optimal evacuation point according to the accident. Acknowledgments This research was supported by BK21 Plus Program, the Climate Change Research Hub of KAIST (Grant No. N11180110) and the Korea Institute of Energy Technology Evaluation and Planning (KETEP) and the Ministry of Trade, Industry & Energy (MOTIE) of the Republic of Korea (No. 20168520021200). Appendix A. Supplementary data Supplementary data related to this article can be found at https://doi.org/10.1016/j.ijnaoe.2018.06.007. References Azzi, C., Pennycott, A., Mermiris, G., Vassalos, D., 2011. Evacuation simulation of shipboard fire scenarios. Fire and Evacuation Modeling Technical Conference 3, 23e29. Cho, Y.O., 2011. Advanced Evacuation Analysis Considering Passenger Behavior in an
15
Emergency. Master Thesis. Department of Naval Architecture and Ocean Engineering, Seoul National University of Korea. Cruise Lines International Association, 2016. Cruise Industry Outlook 2017. Cruise Lines International Association, Washington, DC. Galea, E., Deere, S., Filippidis, L., 2012. The Safeguard Validation Data Setdsgvds1 a Guide to the Data and Validation Procedures. Fire Safety Engineering Group, University of Greenwich. Ginnis, A.I., Kostas, K.V., Politis, C.G., Kaklis, P.D., 2010. VELOS: a VR platform for ship-evacuation analysis. Comput. Aided Des. 42 (11), 1045e1058. Gwynne, S., Galea, E.R., Lyster, C., Glen, I., 2003. Analysing the evacuation procedures employed on a Thames passenger boat using the maritimeEXODUS evacuation model. Fire Technol. 39 (3), 225e246. Ha, S., Cho, Y.O., Ku, N., Lee, K.Y., Roh, M.I., 2013. Passenger ship evacuation simulation considering external forces due to the inclination of damaged ship. Journal of the Society of Naval Architects of Korea 50 (3), 175e181. Hart, P.E., Nilsson, N.J., Raphael, B., 1968. A formal basic for the heuristic determination of minimum cost paths. IEEE Trans. Syst. Sci. Cybern. 4 (2), 100e107. IMO, 1999. MSC.1/Circ. 909. Interim Guidelines for a Simplified Evacuation Analysis on Ro-ro Passenger Ships. IMO, 2016. MSC. 1/Circ. 1533. Revised Guidelines on Evacuation Analysis for New and Existing Passenger Ships. International Maritime Organization. Kim, H., Haugen, S., Utne, I.B., 2016. Assessment of accident theories for major accidents focusing on the MV SEWOL disaster: similarities, differences, and discussion for a combined approach. Saf. Sci. 82, 410e420. Kisko, T.M., Francis, R.L., Nobel, C.R., 1998. Evacnet4 User's Guide. University of Florida. KMST, 2014. Investigation Report of the Capsizing of the MV SEWOL (In Korean). Korean Maritime Safety Tribunal(KMST), South Korea. Korhonen, T., Hostikka, S., 2009. Fire Dynamics Simulator with Evacuation: FDSþ Evac. Technical Reference and User’s Guide. VTT Technical Research Centre of Finland. Lee, D., Kim, H., Park, J.H., Park, B.J., 2003. The current status and future issues in human evacuation from ships. Saf. Sci. 41 (10), 861e876. MIXAMO, 2017. Online [Cited : October 16, 2017]. https://www.mixamo.com. Nelson, H.E., Mowrer, F.W., 2002. In: DiNenno, P., Walton, D.W. (Eds.), Emergency Movement, the SFPE Handbook of Fire Protection Engineering. National Fire Protection Association. Oh, S.G., 2017. Personal Discussion. MBC, PD Note, 1124. Online [Cited : April 18, 2017.] URL. https://www.youtube.com/watch?v¼V98tgWMBtJ0. Park, H.J., Lee, Y.J., 2016. A study on the standardization of ASET on deck of ro-ro passenger ship through simulation analysis of the SEWOL ship turnover accident. Journal of Korean Society of Hazard Mitigation 16 (2), 495e503. Pradillon, J.Y., 2004. ODIGO-modelling and simulating crowd movement onboard ships. In: 3rd International Conference on Computer and it Applications in the Maritime Industries, COMPIT, Siguenza, Spain, pp278-289. Reynolds, C.W., 1987. Flocks, herds and schools: a distributed behavioral model. In: ACM SIGGRAPH Computer Graphics (Vol. 21, No. 4, pp. 25e34). ACM. Ronchi, E., Reneke, P.A., Peacock, R.D., 2014. A method for the analysis of behavioural uncertainty in evacuation modelling. Fire Technol. 50 (6), 1545e1571. Roth, S.D., 1982. Ray casting for modeling solids. Comput. Graph. Image Process. 18 (2), 109e144. Santos, G., Aguirre, B.E., 2004. A Critical Review of Emergency Evacuation Simulation Models. Snape, J., Guy, S.J., Vembar, D., Lake, A., Lin, M.C., Manocha, D., 2012, March. Reciprocal collision avoidance and navigation for video games. In: Game Developers Conf., San Francisco. Togawa, K., 1955. Study of Fire Escape Based on the Observation Multitude Currents. Japan Building Research Institute. Report 55e14. Unity, 2008. Unity game Engine-official Site. Online [Cited : October 9, 2008.]. http://unity3d.com. Vassalos, D., Kim, H.S., Christiansen, G., Majumder, J., 2002. A Mesoscopic Model for Passenger Evacuation in a Virtual Ship-sea Environment and Performancebased Evaluation. Zhang, D., Shao, N., Tang, Y., 2017, May. An evacuation model considering human behavior. In: Networking, Sensing and Control (ICNSC), 2017 IEEE 14th International Conference on (pp. 54e59). IEEE.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
Contents lists available at ScienceDirect
International Journal of Naval Architecture and Ocean Engineering journal homepage: http://www.journals.elsevier.com/ international-journal-of-naval-architecture-and-ocean-engineering/
Passenger evacuation simulation considering the heeling angle change during sinking Hyuncheol Kim a, Myung-Il Roh b, Soonhung Han a, * a b
Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology, Daejeon, South Korea Department of Naval Architecture & Ocean Engineering, Seoul National University, Seoul, South Korea
a r t i c l e i n f o
a b s t r a c t
Article history: Received 28 October 2017 Received in revised form 4 June 2018 Accepted 25 June 2018 Available online xxx
In order to simulate the evacuation simulation of a ship during a sinking, the slope angle change of the ship must be reflected during the simulation. In this study, the passenger evacuation simulation is implemented by continuously applying the heeling angle change during sinking. To reflect crowd behavior, the human density and the congestion algorithm were developed in this research and the walking speed experiment in the special situation occurring in the inclined ship was conducted. Evacuation simulation was carried out by applying the experimental results and the change of the walking speed according to the heeling angle of the ship. In order to verify the evacuation simulation, test items suggested by International Maritime Organization (IMO) and SAFEGUARD Validation Data Set conducted on a large Ro-PAX ferry (SGVDS 1) which performed real evacuation trial in full-scale ships were performed and the results of simulation were analyzed. Based on hypothetical scenario of when a normal evacuation command is delivered to the passengers of MV SEWOL in time, we predicted and analyzed the evacuation process and the number of casualties. © 2018 Production and hosting by Elsevier B.V. on behalf of Society of Naval Architects of Korea. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Keywords: Advanced evacuation analysis Heeling angle change Walking speed experiment MV SEWOL UNITY
1. Introduction Cruise passengers are expected to total 25.8 million in 2017 and have increased by 62% over the past decade (CLIA, 2016). As demand for cruise and passenger ships is expected to increase steadily, safe ship design and predictions of disaster situations are essential. In the case of these cruises or passenger ships operating at sea, there are usually many people on board and their evacuation route is narrow and complicated. In addition, when the sinking occurs, the slope is changed and the evacuation time is limited. Also, due to the difficulties in rescue, it can cause serious casualties in the event of a disaster. Therefore, accurate evacuation analysis is needed that reflects the special environment of marine environment that is stricter than the buildings onshore. IMO MSC.1/Circ. 1533 (IMO, 2016) proposes a method of using computer simulation that reflects the characteristics of individual passengers. Although it is not mandatory, the simulation result is recognized as a recommendation item and safety of passengers is secured by calculating the time required for the passenger to evacuate in an emergency situation and
* Corresponding author. E-mail address: [email protected] (S. Han). Peer review under responsibility of Society of Naval Architects of Korea.
identifying and eliminating the congestion and risk factors. In this research, to accurately reflect the sinking situation on simulation, the heeling angle change with time was applied continuously. Human density algorithm and congestion algorithm developed in this study were applied to reflect crowd behavior. Also, we applied the experimental results on the walking speed in a specific situation occurring in an inclined ship. For the verification of the simulation, the test items proposed by the IMO and the Safeguard validation data set-SGVDS1, which is an experiment on the actual ship (Galea et al., 2012), were performed and simulation results were analyzed whether the results met the criteria. By using the developed techniques in this research, the evacuation simulation for the scenario of three angles was performed and analyzed for MV SEWOL. In order to compare and analyze the evacuation status and time according to the heeling angle, the simulation was performed about the situation when the accident is not occurred at a heeling angle of 0 , the heeling angle of 30 about 8:50 AM and finally the heeling angle of 52.2 about 9:34 AM. Using the simulation results, we predicted and analyzed the evacuation process and the number of casualties based on the hypothetical scenarios when the normal escape command was delivered in time. Applying the processes and results of this research to other passenger ships, it is possible to design a ship with high reliability of safety by
https://doi.org/10.1016/j.ijnaoe.2018.06.007 2092-6782/© 2018 Production and hosting by Elsevier B.V. on behalf of Society of Naval Architects of Korea. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
2
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
confirming the evacuation possibility of passengers for various disaster situations and modifying and supplementing the congestion point. In addition, we expect to be able to minimize the casualties in case of an actual emergency situation by understanding the optimum measures for the accident situation in advance. 2. Related works The evacuation simulations for ships were mainly carried out since the early 2000s, following the study of buildings onshore. The most important factor in the evacuation simulation is to express the behavior of the passenger and the external environments that affect passengers. Walking speed is one of the most important factors in passenger behavior. Walking speed depends on passenger's age, gender, height, weight, size, disability, destination, environment, risk factors, awareness time, etc. The simulation results vary greatly depending on walking speed. Agent-based method is able to reflect individual characteristics. The agent-based method uses the coordinate system to move passengers and thus it is possible to indicate the exact position of each passenger. However, agent-based method requires immense computational capacity and time. In recent years, an activity-based method has been used in combination with other existing methods to express behaviors in specific situations (Santos and Aguirre, 2004). It is also important to reflect not only individual behavior, but also crowd behavior and emergency behavior in evacuation simulation that large numbers of people move. Cohesion behavior to move to a group location, Alignment behavior to move along the direction that the group moves and separation behavior to keep a certain distance between passengers in a group are considered as representative group behavior in evacuation simulation (Reynolds, 1987). Table 1 shows a comparison of representative methods and features of ship evacuation simulation researches that have been performed so far. Evi is a representative simulation software of passenger ship evacuation analysis that expresses cohesion and alignment behaviors by moving at the same walking speed between groups. The separation behavior is expressed by a method of calculating the number of passengers in a certain space and reducing the walking speed when the density is high (Vassalos et al., 2002). FDS þ EVAC is specialized for smoke and fire propagation and is mainly used for evacuation analysis in case of fire in a closed space such as a building or a turnnel (Korhonen and Hostikka, 2009). maritime EXODUS is a passenger evacuation analysis program using velocity based fine network model. The passenger uses each cell that stored the distance from the starting point to the arrival point to move in the direction of decreasing the distance value (Gwynne et al., 2003). Pathfinder has a SFPE mode that reflects the criteria presented by the U.S Fire Technology Society. However, it does not reflect the situation in which slope of ship is continuously changed as other softwares.
ODIGO is a program for marine specific, using artificial intelligence to express the behavior of passengers (Pradillon, 2004). The simulation of this study was implemented using the UNITY engine. UNITY engine is a game development platform that adds Graphical User Interface (GUI) to Application Programming Interface (API). It has high visualization quality and freedom of development, and can describe various physical characteristics and environment (UNITY, 2008). The previous researches that applied equally reduced walking speed during simulation by applying the fixed slope at a certain time point despite the importance in the ship evacuation simulation. In this research, we apply the change of slope to the passenger continuously by recalculating the walking reduction factor according to the angle during the simulation. We also applied the human density and congestion algorithm developed in this research to simulation to reflect human characteristics more realistically. One of the most difficult parts of the evacuation simulation is verification and validation of the results. Not only is there not enough data for verification, but it is not possible to experiment with the actual scale of an emergency situation that has to escape, so in many cases it depends on assumptions and conjectures to analyze the simulation results. Using the MIXAMO Engine (MIXAMO, 2017) that provides skeletons and animations to 3D models, realistic motion is visualized for passengers. It will be helpful to understand and evaluate the evacuation situation in the result analysis. 3. Modeling method and algorithm For high accuracy evacuation simulations, it is important to use algorithms that reflect actual human characteristics and describe detailed disaster environment. This chapter introduces the techniques applied in this paper, describes the implementation method and the applied algorithm. 3.1. Walking speed reduction factor due to slope of a ship Walking speed is one of the most important factors in evacuation simulation. Passengers are greatly influenced by individual characteristics, crowd density and walking environment. In the case of a ship, passengers may have to evacuate from an inclined ship as well as the normal situation. This situation make the movement more difficult than general situation and causes a walking speed reduction and greatly lengthens the total evacuation time of passengers. Therefore, it is necessary to consider the walking speed change according to the slope when performing the evacuation simulation about the ship. Typically, it is assumed that a person cannot walk at the slope greater than 35 due to the loss of friction. Therefore, most of the studies have been performed only at the slope below 35 (Cho, 2011). Lee et al. (2003) compared and analyzed the walking speed experiments when the trim and
Table 1 Technical comparison of previous works with our work. Gwynne et al. (2003) Ginnis et al. (2010)
Azzi et al. (2011) Park and Lee (2016) Ha et al. (2013) Zhang et al. (2017) This work
VELOS
Evi
Pathfinder
In-House
FDS þ EVAC
UNITY
Agent model Agent rotation Account for slope
maritime EXODUS Velocity X Constant
Velocity B Constant
Velocity X Constant
Velocity B Constant
Acceleration B Constant
Acceleration B None
Method considering slope
Reduction factor
Projection of gravity force Reduction factor Reduction factor
External force X △
Reduction factor X △
Velocity B Continuous/ Prediction equations Reduction factor B B
Simulation platform
Account for walking direction X Visualization △
X B
X △
X B
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
3
heeling angle were present. Cho (2011) applied a regression analysis to the walking experiment results to find the walking speed reduction factors according to slope and the reduction factors are obtained about three situations where the flat ground is inclined and five situations where the stair is inclined. In this research, the simulation was implemented using the results of regression analysis on the walking test results of Cho (2011). Fig. 1 shows each situation. Fig. 2 is the plot of each estimation equation about reduction factor. At 0 without slope, the reduction factor is 1 and there is no decrease in walking speed due to the slope. However, the reduction factor reaches 0 because it is assumed that walking is impossible at the slope greater than 35 . Also, reduction factors for all situations is equal to 0 at 1 and 35 at 0, but the reduction factors according to each situation are different even though the slope is same at the angle between 0 and 35 . In this research, evacuation simulation was performed using the reduction factor to apply the walking speed change according to the slope of the ship. Fig. 2. Plots of walking speed reduction factor.
3.2. Walking speed experiment in corridor of inclined ship In a situation where there is a slope like a ship, it is possible to increase the accuracy of the simulation by applying the walking speed in special situation as well as the walking speed change due to the slope. When there is no slope, the corridors and walls turn into walls and corridors when the slope is present, respectively. A passenger may have to walk using a wall that has changed from wall to corridor due to the slope. In order to apply the walking speed for this special situation to the simulation, a simple experiment was conducted on two 24 years old women using a truck capable of creating an inclined environment (Oh, 2017). Fig. 3 represents the experimental scene. Experiments were conducted on four situations, such as walking with one leg on the floor and another on the wall (walking with wall and floors), rolling sideways on a floor or wall (rolling
sideways), walking using the corner between the floor and the wall (corner walking), and moving while putting two feet on the floor and hands on the wall (walking sideways with hands and feet) that can occur in a situation when the heeling angle is present. Table 2 shows the experimental results for four situations. In IMO MSC.1/Circ. 1533, the minimum walking speed of females younger than 30 years in flat terrain is 0.93 m/s and in stairs up and stairs down are 0.47 m/s and 0.56 m/s, respectively. The minimum speed while climbing stairs is similar to the speed while walking with one leg on the floor and another on the wall (walking on wall and floor). The consideration here is that forces received by a human body due to posture while walking on wall and floor, is similar to the one while climbing stairs. In case of rolling sideways, walking sideways and walking through corners, low movement speed was
Fig. 1. Walking cases in inclined ship.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
4
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 3. Experiment on special situations.
Table 2 Experiment results on special situations.
Moving speed (m/s)
Walking with wall and floors
Roll sideways
Corner walking
Side walking using hands and feet
0.469
0.347
0.303
0.322
measured, since such movements are difficult comparing to regular ones. This experiment is intended to provide an approximate estimation of the walking speed in special situations and has limitations with too few participants. In order to achieve more accurate results, we will conduct experiments later with varying gender and age group.
3.3. Heeling angle change during sinking The ship that is sinking changes its slope according to time. In order to simulate the accurate and realistic evacuation of the passenger ship, the evacuation simulation should be performed by applying the slope varying with time to the walking speed of passenger instead of applying fixed walking speed according to the slope at a specific time. Passenger ships are more complex than buildings, so passengers will go through all route states of the transverse ground and the upward and downward ground during the evacuation. Even though the slope is the same, the walking speed reduction factor varies depending on the route state. Therefore, it is necessary to know the current route status of the passenger in order to apply the correct reduction factor to the evacuation simulation. Fig. 4 shows the walking situation that can be occurred in a ship having a heeling angle. Since the simulation proceeds according to the discrete time interval, it is possible to obtain dyp by comparing yp n that position of current time step and yp n1 that the position of previous time step in the y-axis coordinate. By using dyp , the current path condition can be recognized. Eq. (1) represents the discriminant used in this study to determine the path condition.
yp n yp n1 ¼ dyp
8 < ¼ 0 : Transversal > 0 : Upward : < 0 : Downward
(1)
If dyp that is difference between yp n and yp n1 is 0, it means that the transversal ground is used because y-axis position does not change. If dyp is larger than 0 or smaller than 0, it means that the upward or downward ground is used, respectively. The evacuation simulation performed in this research is agent-based method and agent-based method represents the position of passengers in a coordinate system. Using the algorithm proposed in this research, it is possible to easily recognize the current path condition and apply the walking speed reduction factor according to the slope and path conditions of the ship.
3.4. Human density algorithm The walking speed of passengers depends on the number of nearby passengers. The walking speed in the case in which the passenger density is small and the passengers do not influence each other is referred to as a free walking speed, and in case of a crowd, the walking speed is slower than the free walking speed due to the influence of other passengers. As the number of passengers nearby increases, i.e., as the passenger density increases, the walking speed decreases. This phenomenon is due to the separation behavior to maintain a certain interval between passengers during the group movement. In the evacuation simulation, it is practical to carry out the simulation by taking into consideration both the walking speed when the group moves instead of free walking speed, and the
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
5
Fig. 4. Route state in a ship with heeling angle.
walking speed when encountering other passengers coming from the opposite direction. At this time, the factor that affects the walking speed is the passenger density, which refers to the number of passengers per area. Lee et al. (2003) compared and analyzed the results of walking speed experiment according to population density. Nelson and Mowrer (2002) reported that a free walking speed is possible until 0.54 person=m2 but walking speed at 3.8 person=m2 or more is significantly slower. In the research of Togawa (1955), the walking speed was reduced to about 0.1 m=s by 12 times when the group moves horizontally at 6.0 person=m2 or more. In IMO MSC/Circ.909 (IMO, 1999), the walking speed according to passenger density is divided into low, optimum, moderate and crush. EVACNET4 (Kisko et al., 1998), an evacuation simulation software, divides the passenger density into six stages according to unit area per person (A: More than 1.2, B: 0.93e1.2, C: 0.65e0.93 D: 0.25e0.65, E: 0.19e0.28, F: Lower than 0.19) and reduces the speed according to stages. However, when calculating the density of human to calculate walking speed, it is not realistic to calculate the occupancy of passengers in a certain space. Instead, when calculating the passenger's walking speed, it is more reasonable to consider only the other passengers who enters the field of view and directly affected the movement of the passenger in which the calculation takes place and ignores the passengers behind the moving direction. In this study, Ray-casting (Roth, 1982) method was used to consider only the passengers directly affecting the movement in density calculation. Ray-casting is a method of projecting rays to cross-check with a specific object. Table 3 shows the comparison of the passenger density method between area occupancy method and this research method. The results of Togawa (1955) and a quadratic function were used to estimate the walking speed reduction factor. Fig. 5 is a plot of the estimation equation used in this research. The minimum walking speed reduction factor is given as 0.08 that is 12 times lower than the speed reduction factor when free walking. During the simulation, the presence of nearby passengers affecting the walking speed is checked at every discretized time step and if there are passengers, the number of passengers is counted. The walking speed reduction factor is calculated from the number of passengers and applied to the walking speed individually set at the start of the simulation to move the passenger at a reduced walking speed according to the number of nearby passengers. In order to examine the feasibility of the proposed human density algorithm in this research, the method without considering human density, the area occupancy method, and a human density algorithm that proposed this research for 100 passengers were compared. Fig. 6 shows the results of these simulations according to time. The maximum walking speed of all passengers was determined by arbitrarily distributing a value of males 30e50 years old
(0.97e1.62 m/s), suggested by IMO MSC.1/Circ. 1533. The result when the passenger density was not taken into consideration showed an unrealistic phenomenon that cannot occur when moving the group. This is only possible when passengers have been trained in advance. Furthermore, walking speed without considering human density did not take into account the influence of other people, which did not reflect the separation phenomenon to maintain a constant spacing among the crowd behavior. In the case of the area occupancy method, the influence of passengers located behind and beside the calculating passenger is considered. Therefore, an excessive walking speed reduction occurs more often than necessary. As time passes, the gap between the passengers who are located in front of the moving direction, who are less affected by other passengers, and the passengers located at the back becomes wider. An unrealistic phenomenon has occurred in which the passengers who are located at the rear are moved together. In addition, the passengers who are performing the calculation show unreasonable results because they consider the passengers behind the moving direction, and move at a reduced speed. The walking speed of this research differs greatly between the front and the rear passengers in the same way as the area occupancy method at the beginning of the simulation. However, as the simulation progresses, the gap is gradually reduced and it is confirmed that the distance between passengers is maintained without moving together. This study confirms that passengers are affected by other passengers within the cognitive area and the crowd behavior varies according to the range and characteristics of the cognitive area. It is more realistic to consider only a certain range of passengers entering the field of vision rather than considering all passengers in a particular area including the rear of the moving direction. 3.5. Path finding & congestion algorithm In this evacuation simulation, the calculation of path finding was performed using the A* algorithm (Hart et al., 1968) and the collision avoidance between passengers was performed using the Reciprocal Velocity Obstacles (RVO) (Snape et al., 2012) model. The A* algorithm is a graph/tree search algorithm that finds the shortest path from a given point to final destination. The cost is estimated by evaluating the heuristic estimate. Eq. (2) shows the cost estimation method of the A* algorithm.
f ðnÞ ¼ gðnÞ þ hðnÞ
(2)
f is the total cost, expressed as the sum of g and h, and n is the point of departure. g represents the moving cost from the starting point to the moving point, and the value of g increases as the calculation proceeds. h is a Heuristic function which means the moving cost from the moving point to the final destination. The Manhattan
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
6
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 3 Comparison between area occupancy and this work method.
Fig. 5. Walking speed reduction factor for varying calculated no. of neighbor agent.
method is used in which the obstacle and the diagonal direction are ignored and only the lateral and transverse direction is considered. Fig. 7 shows the process of reaching and calculating the arrival cost from the starting point to the destination using the A * algorithm. The RVO model was developed to overcome the vibration phenomenon of the Agent, which is a disadvantage of the Velocity Obstacle (VO) model, and to give the agent more realistic behavior. In the VO model, agents that are computed assuming that others
are obstacles are responsible for all collision avoidance. In the RVO, unlike the VO model, it is assumed that the other person is not the obstacle and all nearby agents are responsible for collision avoidance. Also, RVO can represent smoother motion than VO model. In addition, in this research, the characteristics of human was reflected at congestion points to enhance reality. To reflect human characteristics as realistically as possible, we analyzed and defined human thought patterns of congestion points. Preferentially, a step to recognize the congestion is needed. It is assumed that even though the same situation is given according to the personality and tendency of an individual, it is different for each person to judge and perceive as a congestion. Once the congestion point is identified, it is assumed that passengers will think or confirm about other available routes and rethink about whether to actually change the route or keep the current route even if other available routes are checked. In order to apply human characteristics to evacuation simulation algorithms, we define four variables (CA , CT , CD and CW ) representing human tendency were defined. Fig. 8 shows the algorithm at the congestion point applied to the evacuation simulation of this research. CA is the congestion agent number that represents the number of passengers existing within a specific forward range of passengers, for whom the calculation is being performed. CT is the congestion thinking number. Each passenger is assigned a different CT value, which represents the tendency of a person. In general,
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
7
Fig. 6. Comparison of simulation results according to density algorithm.
Fig. 7. Process of A star algorithm for path finding.
Fig. 8. Logic of congestion algorithm in simulation.
people show different tendencies, even though they are in the same condition. Only when C_A > C_T, a passenger recognizes that there is a congestion. The congestion decision number, C_D, and congestion waiting number, C_W, are related to the changing of route at the last stage. Each passenger is also assigned a different C_D and C_W value, as they are related to the personality of a passenger. We assume that, even if the passengers are aware that the current path is congested, they will also be reluctant to change the path. Only when C_D > C_W, the passengers will actually change the route and proceed to evacuate. In order to change the route during the simulation, first of all, it is necessary to exclude route recognized to be congestion and re-search the route. This process is implemented by using bit mask. The bit mask is to assign a mask such as 0 or 1 to a bit. Each corridor is assigned an eigenvalue increasing by 2n and the decimal value obtained by using route discriminant shown in Eq. (3) is converted into binary numbers. Each passenger can recognize the congestion
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
8
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 4 Eigenvalue according to corridor number. Bit
∙∙∙
LSB-5
LSB-4
LSB-3
LSB-2
LSB-1
LSB
Corridor number
∙∙∙
5
4
3
2
1
Eigen value
2n
32
16
8
4
2
Route State 0 or 1
phenomenon of the route through the bit mask. Table 4 shows the eigen value and bit according to the route. The Least-Significant Bit (LSB) indicates the least-significant bit that located at the end of the bit configuration. An example of some cases is shown in Table 5. Route discriminant ¼ Path state þ Eigen value
(3)
If Routes 1, 3, and 5 are detected as congestion corridors, this will result in a decimal value of 43 by the route discriminant equation, and 101011 when converted to binary numbers. Conversely, the first bit indicates the existence of the route itself, and the second bit indicates the corridor number in order. A value of 1 in the route indicates that the route is a congestion and should be changed and a value of 0 indicates a route that is not a congestion or an unidentified route. When each occupant judges that the current route is unavailable due to the congestion, the route re-search must be performed again, and the evacuation should proceed using the other route. It is possible to search for the route again, except for the unavailable routes that are recognized by using the bit mask. Fig. 9 shows the simulation results for the behavior of the passengers at the congestion point. 4. Validation/verification of evacuation simulation IMO established MSC.1/Circ. 1533 to acknowledge the evacuation time derived from computer simulations. In addition, guidance on validation/verification of evacuation simulation tools was presented in ANNEX 3, and the tests for verifying whether a reasonable evacuation simulation can be performed is presented. Tests 1e7 are component testing that involves checking that the various components of the software perform as intended, and tests 8e12 are qualitative verification that concerns the nature of predicted human behavior with informed expectations. The evacuation simulation implemented in this research performed the tests presented in IMO MSC.1/Circ. 1533 for validation and verification. It was confirmed that all 12 tests requirements were satisfied. In this chapter, to demonstrate that the test requirements are satisfied, the results of tests 8, 10, and 12 that are more complex qualitative verification tests rather than the relatively simple component tests, are presented. In addition to the IMO test, the SAFEGUARD validation data set 1 was used for further verification and the results were analyzed. 4.1. IMO test 4.1.1. Test 8: counterflow Test 8 takes place in a space where two rooms, 10 m wide and long, are connected via a corridor 2 m wide and 10 m long. In the first room, 100 passengers move through the corridor to the second Table 5 Example of route status. Unavailable route
Route discriminant
1 5 1,3,5 2,4,6
1 1 1 1
þ þ þ þ
2¼3 32 ¼ 33 2þ8 þ 32 ¼ 43 4þ16 þ 64 ¼ 85
Binary number 11 100001 101011 1010101
Fig. 9. Simulation of congestion (MV SEWOL A deck).
room. When the passengers in the first room move to the second room, the time when there is no passenger coming from the opposite side and when 10, 50, and 100 passengers are coming from the opposite side are measured. The expected result is that as the number of passengers coming from the opposite side increases, the time required for the passengers in the first room to reach the second room increase. Fig. 10 shows the simulation results for the case where the counterflow is absent, and the case where the counterflow increases at 10, 50, and 100 passengers, respectively. It is confirmed that as the amount of counterflow increases, the time required to move from the first room to the second room increases. Table 6 shows the results of comparison with other research results for additional verification. 4.1.2. Test 10: Exit route allocation Test 10 takes place in a space where a total of 23 passengers are distributed among 12 rooms. The expected result is that the allocated passengers move to the appropriate destinations. Fig. 11 shows the result of test 10. It was confirmed that all the passengers move to the appropriate destinations along the corridor without penetrating the wall. 4.1.3. Test 12: flow density relation Test 12 is for demonstrating that the flow of persons in the corridor is generally smaller at very high population densities
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
9
Fig. 10. Simulation result of test 8.
Table 6 Comparison of simulation results. No. of counterflow
Cho (2011)
Evi
This research
88.9 () 125.6 (41.3%) 229.1 (157.7%) 327.9 (268.8%)
82.3 () 108.1 (31.3%) 224.3 (172.5%) 277.7 (237.4%)
(sec) 0 10 50 100
84.6 () 93.2 (10.2%) 137.1 (62.1%) 216.1 (155.4%)
compared with that at moderate densities. From the same starting point, 20, 50, and 70 passengers were deployed to conduct the test, and the simulation results are shown in Fig. 12. The flow of persons was calculated by measuring the time taken for the last passenger to pass through the 30-m corridor. Table 7 shows the results according to each number of passengers. It was confirmed that as the number of passengers increased, the flow of persons decreased. This phenomenon can be regarded as a result of the separation behavior among the crowd behaviors to maintain the minimum distance between passengers.
4.2. SAFEGUARD validation data set 1 SAFEGUARD Validation Data Set 1 (SGVDS 1) was conducted on an actual Large RO-PAX ferry, with a total of 1349 passengers
boarded. There are four metrics to validate the simulation. The first is Euclidean Relative Difference (ERD), as shown in Eq. (4). This is used to assess the distance between the experimental data ðEi Þ and the model data ðmi Þ. If ERD is 0, this means that the two data are identical.
Em ¼ jEj
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1 ðEi mi Þ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Pn 2 i¼1 Ei
(4)
The second is Euclidean projection coefficient (EPC), as shown in Eq. (5). The EPC calculates a factor which when multiplied by each model data point ðmi Þ reduces the distance between the model ðmÞ and experimental ðEÞ vectors to its minimum. If EPC is 1.0, it implies that the difference between the model ðmÞand experiment ðEÞ data is as small as is possible.
Pn E; m Ei mi ¼ Pi¼1 n 2 2 m i¼1 mi
(5)
The third is the Secant cosine, as shown in Eq. (6). It provides a measure of how well the shape of the model data curve matches that of the experimental data curve. If SC is 1.0, it means that the shape of the model ðmÞ curve is identical to that of the experimental ðEÞ curve.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
10
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 11. Simulation result of test 10.
Pn
ðEi Eis Þðmi mis Þ i¼sþ1 s2 ðti ti1 Þ
E; m ¼ rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi Em Pn ðEi Eis Þ2 Pn ðmi mis Þ2 i¼sþ1 s2 ðti ti1 Þ
(6)
i¼sþ1 s2 ðti ti1 Þ
The last parameter is the percentage difference between the predicted Total assembly time (TAT) and the measured TAT. Eq. (7) indicates TAT.
%TAT ¼ ðMeasured TAT Predicted TATÞ*100 = Measured TAT (7) The ship used for validation consists of three decks and Fig. 13 shows modeling results. According to SGVDS 1 guidelines, a total of 50 simulations were carried out while changing the distribution of the population in every fifth simulation as stated in IMO MSC/Circ. 1533. Table 8 shows Total Assembly Time (TAT) and ERD. The simulation of the smallest ERD should be selected for the basis of the validation
comparison. In this case, the values of case no. 11 are selected for analysis about two phase assessment process. The value calculated for case no. 11 are listed in Table 9. The suggested two phase assessment process in the SGVDS 1 guideline are as follows: - Phase 1: For the predicted total assembly curve, determine ERD, EPC, SC and % TAT. 1. ERD 0.45. 2. 0.6 EPC 1.4. 3. SC 0.6 with s=n ¼ 0.05. 4. Predicted TAT for the overall assembly to be within 45% of the measured value. - Phase 2: for the predicted assembly curve for each of the four assembly stations, determine ERD, EPC and SC. At least 9 out of 12 criteria must be met SGVDS 1 to satisfy the criteria and it is not acceptable to have two or more failed criteria in any one assembly station. The simulation results for SGVDS 1 in this study confirm that both Phase 1 and Phase 2 presented in the guideline are satisfied. 5. MV SEWOL evacuation simulation 5.1. System architecture The simulation in this research is divided into the PreTable 7 Comparison of flow of persons.
Fig. 12. Simulation result of test 12.
No. of agent
Density
Time ðsecÞ
Flow of persons ðm=sÞ
%
ðp=m2 Þ
20 50 70
0.95 2.38 3.33
57.78 94.48 112.28
0.63 0.38 0.33
e 39.68 47.62
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
11
Fig. 13. Geometry model of SGVDS 1 and distribution of passengers.
Table 8 TAT and ERD for SGVDS1 simulation. No.
Time
ERD
No.
Time
ERD
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
440.8 424.1 432.5 403.6 428.7 416.2 440.4 430.6 435.5 438.7 422.3 430.5 411.6 375.2 393.4 440.0 390.3 440.0 390.3 415.2 431.8 437.4 386.3 411.9 393.9
0.392 0.414 0.402 0.447 0.395 0.421 0.403 0.415 0.408 0.412 0.371 0.391 0.399 0.427 0.412 0.389 0.426 0.408 0.418 0.397 0.406 0.399 0.435 0.415 0.402
26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
428.9 411.4 442.3 384.9 413.6 449.3 421.7 435.5 399.3 411.1 405.2 443.4 576.4 509.9 386.6 418.7 402.6 429.2 409.4 419.3 402.6 420.9 411.7 416.3 441.8
0.405 0.403 0.397 0.432 0.414 0.422 0.432 0.427 0.413 0.410 0.411 0.433 0.442 0.436 0.392 0.437 0.414 0.430 0.400 0.414 0.391 0.385 0.388 0.406 0.423
Table 9 Validation metric values of smallest ERD case for SGVDS 1.
Overall AS A AS B AS C AS D
ERD
EPC
SC (s/n ¼ 0.05)
%diff. TAT
0.371 0.335 0.381 0.256 0.668
0.897 1.190 0.909 0.977 0.630
0.877 0.698 0.835 0.820 0.706
27.81 43.28 28.37 22.93 30.10
Processing stage and the Run Evacuation Simulation stage. Fig. 14 shows the whole simulation process. In the Pre-Processing stage, analysis of the sinking of the ship for application to the simulation environment is performed and the initial information necessary for running the simulation is defined. The analysis of the sinking of the ship is transferred to Motion of Ship in the Run Evacuation
Simulation stage. The Definition of Evacuation Simulation is a step that takes all the overall matters and gives the passengers detailed characteristics (gender, age, body shape, athletic ability, personality, etc.) and assigns the initial position and the destination point. It also creates a navigation field for defining the movable space and applying the pathfinding algorithm. Each corridor is assigned an eigenvalue according to the bit in order to allow the passengers to cope with the congestion. When the Pre-Processing stage is completed, the Run Evacuation Simulation stage is executed. In the Global Navigation stage, the route is calculated from the initial position to the destination using the A* algorithm based on the information in the Pre-Processing stage. When the route calculation of all the passengers is completed, the Local Navigation stage is performed. In the Local Navigation stage, collision avoidance is performed between the passengers moving to the destination, and ray-casting is used at every time step to calculate the number of nearby passengers affecting each passenger's walking speed. The congestion algorithm is executed based on the calculated number of passengers. If the congestion condition is satisfied, the route is recalculated and the evacuation proceeds. In Motion of Ship stage, information on the slope from Ship Sinking Analysis stage and the current route condition of each passenger is obtained. In the Walking Speed Reduction Factor stage, the reduction factor is calculated based on the slope and route condition calculated in the previous stage and the number of nearby passengers that affect walking speed. The evacuation simulation is performed by repeating the Local Navigation stage by adding the discretized time until the passenger either is unable to move due to the slope of the ship or has reached the destination point. 5.2. Simulation scenario The simulation of evacuation simulation in MV SEWOL (Length: 145.6 m, Breadth: 22.0 m, Draft: 6.25 m, and Gross Registered Tonnage: 6825) was based on the report of the research institute and academic journal (KMST, 2014; Kim et al., 2016). Fig. 15 indicates the position and number of crews and passengers of MV SEWOL's Nav Deck (5th), A Deck (4th) and B Deck (3rd) according to the reservation record of the day of the accident. A total of 476 people boarded on the day of the accident, 17 on Nav Deck, 336 on A Deck and 123 on B Deck.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
12
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Fig. 14. Process of evacuation simulation.
Each simulation case was performed with awareness time of 5 min randomly assigned to the passengers since the time of the sinking accident occurred during the daytime and passengers start moving immediately without awareness time. The walking speed
of passengers was determined based on table 3.4 in IMO MSC.1/ Circ. 1533. Considering that most of the passengers were high school students, walking speed of man was applied by males younger than 30 years old that maximum walking speed is 1.85 m/s
Fig. 15. Deck layout of MV SEWOL.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
and minimum walking speed is 1.11 m/s. The walking speed of woman was applied by females younger than 30 years old that maximum walking speed is 1.55 m/s and minimum walking speed is 0.93 m/s. All crew members were assumed to be male and maximum walking speed and minimum walking speed are 1.85 m/ s and 0.93 m/s, respectively. The walking speed between the highest walking speed and the lowest walking speed of the corresponding group was randomly assigned to passengers to finally determine the passenger's walking speed in simulation. The recognition of the evacuation location and route was performed on the assumption that it was done in advance. The earliest possible time to make evacuation command is 8:50 AM at a heeling angle of 30 immediately after the accident. Based on an investigation report of the capsizing of the MV SEWOL that the heeling angle was 30 at 8:50 AM and 74 at 10:09 AM, the simulation was performed to reflect the heeling angle change under the assumption that MV SEWOL was linearly inclined as a 0.56 deg=min of heeling angular velocity. Because the stairs are difficult to use in a situation where the heeling angle of the ship is 30 , the tidal current at the accident spot is very fast and the water temperature is low, it was assumed that if there was an evacuation command, it would be more likely to induce the initial evacuation to the starboard deck that is the opposite direction to the sinking rather than to jump directly into the sea. The next possible time to make evacuation command is 9:34 AM at a heeling angle of 52.2 when the coast guard arrived and the crews were aware of the arrival of them. Normally, a person cannot walk at the slope greater than 35 due to the loss of friction. However, in the case of MV SEWOL, there were passengers who succeeded in evacuation at a slope of 52.2 or more. Therefore, evacuation simulations were performed under the assumption that passengers can grab structures and slip down. Since there is no experimental and simulation results for the slope of more than 35 , the walking speed was simulated by applying the mobility impaired (2), which is the slowest speed suggested by IMO. In the case of severe slope of the ship, the corridor becomes the wall and the wall becomes the corridor. We carried out the simulation using the experimental result of walking with wall and floors mentioned in section 3. In addition, in the situation where the heeling angle of ship is 52.2 , we assumed that it was possible that passengers can grab structures and slip down but it is impossible to climb up. Based on this assumption, simulation was performed by setting the port side that is the inclined direction as the evacuation point. To compare the simulation results of evacuation time, simulation that heeling angle of ship is 0 were also conducted to evacuate to the port and starboard side of each deck.
5.3. Evacuation simulation Table 10 shows the scenario and results of the evacuation simulation performed in this research and in Fig. 16, overall simulations are shown. The simulation video in this research can be found in Appendix A. Supplementary Video. Since the Total Evacuation Time (TET) is the most important result in this research, the results of TET are compared and TET are selected based on the convergence measure of total evacuation time (Ronchi et al., 2014). Eq. (8) represents the average value of TET when the simulation is performed n times and i is the i-th simulation run.
TETave ¼
n 1X TETi n
(8)
i¼1
A measure of the convergence of two consecutive mean total evacuation times is obtained calculating convergence of total evacuation time (TETconv ) and Eq. (9) represents TETconv .
TETave TETavn1 TETconv ¼ TET
13
(9)
ave
In this research, we use 1% of TETconv as our criterion. All simulation cases were repeated 50 times and TETconv of all results was less than 1% and confirmed that it met the criterion. 5.3.1. Case 1 (Heeling angle: 0 deg) Case 1 is simulation for evacuation time comparison. It was performed with a heeling angle of 0 , and the accident did not occur. Because there is no slope, the slope-dependent walking speed reduction factor is fixed to 1 during the simulation, and the simulation was performed only considering the walking speed reduction factor according to human density. When passengers evacuated to the port or starboard side, significant differences were not seen in the results. In the case of B deck, where the number of passengers is less than half that of A deck, the time required for evacuation is relatively long. The reason for this can be predicted as a result of the fact that B deck had many women, who were walking slower than men in the forward rooms that were farthest from the evacuation point. 5.3.2. Case 2 (Heeling angle: 30 deg) Case 2 is a situation that the heeling angle of ship is 30 at 8:50 AM, immediately after the accident based on hypothetical scenario of when evacuation command is delivered. The heeling angle of MV SEWOL increased at an angular velocity of 0.56 /min. It took only 535 s to reach 35 when the walking speed reduction factor becomes 0 due to the slope. Therefore, it is safe to successfully evacuate to the evacuation area in 535 s of ASET. Considering the awareness time, A deck took 439 s to evacuate the whole passengers. B deck was failed to evacuate the whole passengers and 97.5% of B deck passengers took 438 s to evacuate. Considering that after 535 s, movement of passengers was impossible, it was not a situation where passengers can evacuate easily. In the research of Park and Lee (2016), the time required for all passengers to evacuate to the port side of the B deck is 305 s at the heeling angle of 30 . Park and Lee (2016) performed a simulation using a walking speed reduction factor with a fixed heeling angle of 30 rather than reflecting the actual angular velocity during the sinking. For this reason, in comparison with the evacuation simulation of this research, it seems that even though it is more complicated and difficult situation such as more passengers and stairs are used, it took less time to evacuate. 5.3.3. Case 3 (Heeling angle: 52.2 deg) Case 3 is around 9:34 AM, just after the arrival of coast guard. At this time, the MV SEWOL was inclined at 52.2 . At a slope of 35 or more, it is common to assume that a person cannot walk because of the loss of friction. However, in the case of the MV SEWOL, some passengers succeeded in evacuating when the heeling angle was more than 52.2 . Therefore, evacuation simulations were performed under the assumption that passengers can grab structures and slip down. It is assumed that the walking speed is mobility impaired(2) when slipping down, and using the experimental result from walking with the floor and wall, described in Section 3, it is assumed that the walking speed is 0.469 m/s in a narrow corridor. In the case of B and A decks, it is impossible to evacuate through the port side due to the submergence according to heeling angle 12 and 15 min after 9:34 AM, respectively. Therefore, the simulations were performed assuming that the ASET of A and B decks are 15 min and 12 min, respectively. The simulation results of A deck show that 582 s with awareness time and 556 without awareness time. In case of B deck, the simulation results show that 429 s with awareness time and 227 s without awareness time.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
14
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
Table 10 MV SEWOL evacuation simulation scenarios and results. Case
Time (AM)
Initial heeling angle (deg)
Deck
Evacuation point
Heeling angular velocity (deg/min)
Awareness time (sec)
ASET (sec)
RSET (sec)
% of evacuee
1e1 1e2 1e3 1e4 1e5 1e6 1e7 1e8 1e9 1e10 1e11 1e12 2e1 2e2 2e3 2e4 2e5 2e6 3e1 3e2 3e3 3e4 3e5 3e6
e
0
NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd) NAV (5th) A (4th) B (3rd)
Starboard side of each deck
0
300
No limitation
301.7 317.6 319.7 36.4 91.8 87.3 298.5 322.1 331.9 29.4 93.1 96.3 354.9 421.9 421.8 64.7 197.3 175.6 325.9 387.8 355.6 67.8 286.9 259.1
100
08:50
09:34
30
52.2
0
Port side of each deck
300
0
Starboard side of each deck
0.56
300
535
0
Port side of each deck
0
300
0
1200 900 720 1200 900 720
Fig. 16. Simulation examples: (a) Overall view of MV SEWOL (b) Passenger classification (c) forward of A deck (d) Aft of A deck (e) Overall view of A deck (f) Forward of B deck (g) Middle of A deck (h) Slipping motion (i) Middle of B deck (j) Walking corridor with wall and floors in inclined ship (k) Slipping motion in A deck.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007
H. Kim et al. / International Journal of Naval Architecture and Ocean Engineering xxx (2018) 1e15
6. Conclusion In this research, for realistic evacuation simulation of passenger ships, the actual heeling angle change of MV SEWOL during sinking according to time was continuously reflected in the passenger's walking speed, and experimental results for special situations were used in simulation. Human density and congestion algorithms were developed and applied to reflect human characteristics. The simulation was carried out at the time when the evacuation command could be taken based on investigation report of the capsizing of the MV SEWOL. Simulation results show that if there was an immediate evacuation command at the time of the accident, all passengers were able to survive. However, at the time of the arrival of the coast guard, there were deaths, even if the evacuation command was delivered. Through this study, it can be seen that the responsibility and prompt response of the captain and the crew are very important factors which are directly connected with the casualties in case of ship accident. Normally, in the case of a slope of 35 or more, it is impossible for a human to walk. However, there may be situations where a human can walk using other structures or slipping down. It is expected that a more realistic simulation will be possible if data on various situations are acquired through experiments. By applying the implementation process and results of this research to other ships, it is possible to confirm evacuation performance and congestion points in advance and to make more realistic evacuation manual. Through such predictions about disaster situation, we expect to minimize the casualties in the event of an actual emergency. In the future research, we plan to study the evacuation simulation by adding the situation that affects the movement of passengers by moving the furniture or equipment due to the slope of the ship and selecting the optimal evacuation point according to the accident. Acknowledgments This research was supported by BK21 Plus Program, the Climate Change Research Hub of KAIST (Grant No. N11180110) and the Korea Institute of Energy Technology Evaluation and Planning (KETEP) and the Ministry of Trade, Industry & Energy (MOTIE) of the Republic of Korea (No. 20168520021200). Appendix A. Supplementary data Supplementary data related to this article can be found at https://doi.org/10.1016/j.ijnaoe.2018.06.007. References Azzi, C., Pennycott, A., Mermiris, G., Vassalos, D., 2011. Evacuation simulation of shipboard fire scenarios. Fire and Evacuation Modeling Technical Conference 3, 23e29. Cho, Y.O., 2011. Advanced Evacuation Analysis Considering Passenger Behavior in an
15
Emergency. Master Thesis. Department of Naval Architecture and Ocean Engineering, Seoul National University of Korea. Cruise Lines International Association, 2016. Cruise Industry Outlook 2017. Cruise Lines International Association, Washington, DC. Galea, E., Deere, S., Filippidis, L., 2012. The Safeguard Validation Data Setdsgvds1 a Guide to the Data and Validation Procedures. Fire Safety Engineering Group, University of Greenwich. Ginnis, A.I., Kostas, K.V., Politis, C.G., Kaklis, P.D., 2010. VELOS: a VR platform for ship-evacuation analysis. Comput. Aided Des. 42 (11), 1045e1058. Gwynne, S., Galea, E.R., Lyster, C., Glen, I., 2003. Analysing the evacuation procedures employed on a Thames passenger boat using the maritimeEXODUS evacuation model. Fire Technol. 39 (3), 225e246. Ha, S., Cho, Y.O., Ku, N., Lee, K.Y., Roh, M.I., 2013. Passenger ship evacuation simulation considering external forces due to the inclination of damaged ship. Journal of the Society of Naval Architects of Korea 50 (3), 175e181. Hart, P.E., Nilsson, N.J., Raphael, B., 1968. A formal basic for the heuristic determination of minimum cost paths. IEEE Trans. Syst. Sci. Cybern. 4 (2), 100e107. IMO, 1999. MSC.1/Circ. 909. Interim Guidelines for a Simplified Evacuation Analysis on Ro-ro Passenger Ships. IMO, 2016. MSC. 1/Circ. 1533. Revised Guidelines on Evacuation Analysis for New and Existing Passenger Ships. International Maritime Organization. Kim, H., Haugen, S., Utne, I.B., 2016. Assessment of accident theories for major accidents focusing on the MV SEWOL disaster: similarities, differences, and discussion for a combined approach. Saf. Sci. 82, 410e420. Kisko, T.M., Francis, R.L., Nobel, C.R., 1998. Evacnet4 User's Guide. University of Florida. KMST, 2014. Investigation Report of the Capsizing of the MV SEWOL (In Korean). Korean Maritime Safety Tribunal(KMST), South Korea. Korhonen, T., Hostikka, S., 2009. Fire Dynamics Simulator with Evacuation: FDSþ Evac. Technical Reference and User’s Guide. VTT Technical Research Centre of Finland. Lee, D., Kim, H., Park, J.H., Park, B.J., 2003. The current status and future issues in human evacuation from ships. Saf. Sci. 41 (10), 861e876. MIXAMO, 2017. Online [Cited : October 16, 2017]. https://www.mixamo.com. Nelson, H.E., Mowrer, F.W., 2002. In: DiNenno, P., Walton, D.W. (Eds.), Emergency Movement, the SFPE Handbook of Fire Protection Engineering. National Fire Protection Association. Oh, S.G., 2017. Personal Discussion. MBC, PD Note, 1124. Online [Cited : April 18, 2017.] URL. https://www.youtube.com/watch?v¼V98tgWMBtJ0. Park, H.J., Lee, Y.J., 2016. A study on the standardization of ASET on deck of ro-ro passenger ship through simulation analysis of the SEWOL ship turnover accident. Journal of Korean Society of Hazard Mitigation 16 (2), 495e503. Pradillon, J.Y., 2004. ODIGO-modelling and simulating crowd movement onboard ships. In: 3rd International Conference on Computer and it Applications in the Maritime Industries, COMPIT, Siguenza, Spain, pp278-289. Reynolds, C.W., 1987. Flocks, herds and schools: a distributed behavioral model. In: ACM SIGGRAPH Computer Graphics (Vol. 21, No. 4, pp. 25e34). ACM. Ronchi, E., Reneke, P.A., Peacock, R.D., 2014. A method for the analysis of behavioural uncertainty in evacuation modelling. Fire Technol. 50 (6), 1545e1571. Roth, S.D., 1982. Ray casting for modeling solids. Comput. Graph. Image Process. 18 (2), 109e144. Santos, G., Aguirre, B.E., 2004. A Critical Review of Emergency Evacuation Simulation Models. Snape, J., Guy, S.J., Vembar, D., Lake, A., Lin, M.C., Manocha, D., 2012, March. Reciprocal collision avoidance and navigation for video games. In: Game Developers Conf., San Francisco. Togawa, K., 1955. Study of Fire Escape Based on the Observation Multitude Currents. Japan Building Research Institute. Report 55e14. Unity, 2008. Unity game Engine-official Site. Online [Cited : October 9, 2008.]. http://unity3d.com. Vassalos, D., Kim, H.S., Christiansen, G., Majumder, J., 2002. A Mesoscopic Model for Passenger Evacuation in a Virtual Ship-sea Environment and Performancebased Evaluation. Zhang, D., Shao, N., Tang, Y., 2017, May. An evacuation model considering human behavior. In: Networking, Sensing and Control (ICNSC), 2017 IEEE 14th International Conference on (pp. 54e59). IEEE.
Please cite this article in press as: Kim, H., et al., Passenger evacuation simulation considering the heeling angle change during sinking, International Journal of Naval Architecture and Ocean Engineering (2018), https://doi.org/10.1016/j.ijnaoe.2018.06.007